Accelerating Lanczos- and DCT-Based Realistic Sparse Matrix Generation Using CUDA and OpenMP Lanczos ve DCT Tabanli Gerçekçi Seyrek Matris Üretiminin CUDA ve OpenMP ile Hizlandirilmasi
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636670
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: CUDA, DCT, Lanczos, Matrix generation, OpenMP, Signal processing, Sparse matrix
- Ankara Yıldırım Beyazıt Üniversitesi Adresli: Evet
Özet
The discrete cosine transform (DCT) and Lanczos resampling are fundamental signal processing techniques known for preserving spectral and structural characteristics. Sparse matrices are widely used in evaluating numerical algorithms and data-driven methods; however, existing collections are limited, and synthetic generators often fail to capture real-world structural patterns. This need has led to signal processing-based approaches such as MatGen, whose sequential Python implementation incurs significant computational overhead. In this work, we present a high-performance C implementation of Lanczos- and DCT-based sparse matrix generation methods in MatGen, accelerated using OpenMP and CUDA. The proposed approach operates directly on sparse representations, improving efficiency while avoiding dense expansions. Experimental results demonstrate substantial performance gains and strong scalability for large-scale sparse matrices.